Auto gating update

April 11th, 2017

OpenCyto

Finak, Frelinger, Jiang, et al. (2014)

  • Mimics manual gating by focusing on 2 channels at a time
    • can follow traditional gate hierarchy
    • e.g. gate lymph, then single, then live, etc
  • Pipeline templates defined in .csv file
    • defines algorithmic approach for each gate to be applied across many samples
    • not “just push go”, takes some setup for a decent template
  • Results are interpretable and labelled populations
    • not geared toward detecting novel cell types

OpenCyto

The two top performing gating algorithms - OpenCyto (v. 1.7.4), flowDensity (v. 1.4.0) - in a study run by the FlowCAP consortium aimed at selecting the best performing algorithms for this larger study were chosen for the analysis presented in this paper. Standardizing Flow Cytometry Immunophenotyping Analysis from the Human ImmunoPhenotyping Consortium Finak, Langweiler, Jaimes, et al. (2016)

  • Stole template from above as a start
    • our panel 1 template currently lives here

Current progress overview

plot of chunk unnamed-chunk-1

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 2265 0.9846 1.015 0.0015 0.9949 0.9972

plot of chunk unnamed-chunk-3

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 2265 0.979 0.9861 0.0031 0.9782 0.989

Example OpenCyto gates

Next few slides

  • For each gate
    • a good-ish example of an auto gate
    • a bad example/common problems
    • overview of gate performance

Lymphocytes (SSC-A v FSC-A)

FN TN TP FP ACC SENS SPEC PREC
Freq 1490 142572 28420 0 0.9914 0.9502 1 1

FN FP TN TP ACC SENS SPEC PREC
Freq 2849 22579 778949 38211 0.9698 0.9306 0.9718 0.6286
  • may decrease the quantile of lymph cluster from 95% back to 90%

Lymphocytes (SSC-A v FSC-A)

plot of chunk unnamed-chunk-7

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9885 1.014 0.0078 0.9912 0.9953

plot of chunk unnamed-chunk-9

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9968 1.038 0.005 0.9966 0.9972

Single Cells (FSC-H v FSC-W)

FN FP TN TP ACC SENS SPEC PREC
Freq 224 10743 1096246 84062 0.9908 0.9973 0.9903 0.8867

FN FP TN TP ACC SENS SPEC PREC
Freq 10397 46 1237576 143142 0.9925 0.9323 1 0.9997
  • Force FSC-W+ to have a minimum cutoff or make more inclusive ? What is the manual strategy?

Single Cells (FSC-H v FSC-W)

plot of chunk unnamed-chunk-13

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9822 1.018 0.0087 0.9892 0.9935

plot of chunk unnamed-chunk-15

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.0585 0.032 0.0225 0.0134 -0.0278

Live cells (PE-)

FN FP TN TP ACC SENS SPEC PREC
Freq 213 10668 1096940 83454 0.9909 0.9975 0.9904 0.8867

FN FP TN TP ACC SENS SPEC PREC
Freq 17050 7378 476522 178534 0.964 0.9128 0.9848 0.9603
  • victim of singlet gate

Live cells (PE-)

plot of chunk unnamed-chunk-19

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9821 1.019 0.0088 0.9889 0.9933

plot of chunk unnamed-chunk-21

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.6801 1.064 0.0931 0.4672 0.6227

Tcells (CD3+ CD19-)

FN FP TN TP ACC SENS SPEC PREC
Freq 4244 273 1454088 136130 0.9972 0.9698 0.9998 0.998

FN FP TN TP ACC SENS SPEC PREC
Freq 1393 3028 603276 43658 0.9932 0.9691 0.995 0.9351
  • can trim sides (and a bit less on left side)

Tcells (CD3+ CD19-)

plot of chunk unnamed-chunk-25

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9958 0.9979 0.0054 0.9956 0.9978

plot of chunk unnamed-chunk-27

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9784 0.9059 0.0133 0.9687 0.9711

Helper Tcells-CD4+

FN FP TN TP ACC SENS SPEC PREC
Freq 880 104 524259 6874 0.9982 0.8865 0.9998 0.9851

FN FP TN TP ACC SENS SPEC PREC
Freq 5807 7577 963616 94294 0.9875 0.942 0.9922 0.9256

Helper Tcells-CD4+

plot of chunk unnamed-chunk-31

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9891 1.015 0.0113 0.982 0.9897

plot of chunk unnamed-chunk-33

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9709 0.9173 0.0215 0.9241 0.9539

central memory helper Tcells (CCR7+ CD45RA-)

central memory helper Tcells (CCR7+ CD45RA-)

plot of chunk unnamed-chunk-35

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9643 0.9344 0.0238 0.9121 0.9542

plot of chunk unnamed-chunk-37

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.898 0.7929 0.0329 0.7959 0.8867

effector helper Tcells (CCR7- CD45RA+)

FN FP TN TP ACC SENS SPEC PREC
Freq 3307 1114 2113444 22741 0.9979 0.873 0.9995 0.9533

FN FP TN TP ACC SENS SPEC PREC
Freq 953 19155 786225 21460 0.9757 0.9575 0.9762 0.5284

effector helper Tcells (CCR7- CD45RA+)

plot of chunk unnamed-chunk-41

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.7592 0.8319 0.0386 0.7567 0.8635

plot of chunk unnamed-chunk-43

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.7198 0.7991 0.0695 0.47 0.661

effector memory helper Tcells (CCR7- CD45RA-)

FN FP TN TP ACC SENS SPEC PREC
Freq 1509 240 1382591 6821 0.9987 0.8188 0.9998 0.966

FN FP TN TP ACC SENS SPEC PREC
Freq 974 6718 948445 12272 0.9921 0.9265 0.993 0.6462

effector memory helper Tcells (CCR7- CD45RA-)

plot of chunk unnamed-chunk-47

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.8323 0.7918 0.0357 0.7677 0.8681

plot of chunk unnamed-chunk-49

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.7935 0.7057 0.0462 0.6104 0.7736

naive helper Tcells (CCR7+ CD45RA+)

FN FP TN TP ACC SENS SPEC PREC
Freq 827 244 926060 26892 0.9989 0.9702 0.9997 0.991

FN FP TN TP ACC SENS SPEC PREC
Freq 18678 251 789709 19155 0.9771 0.5063 0.9997 0.9871

naive helper Tcells (CCR7+ CD45RA+)

plot of chunk unnamed-chunk-53

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9714 0.9777 0.0133 0.9733 0.9853

plot of chunk unnamed-chunk-55

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9394 0.7853 0.0327 0.7947 0.885

cytotoxic Tcells-CD8+

FN FP TN TP ACC SENS SPEC PREC
Freq 2478 2311 1665944 30763 0.9972 0.9255 0.9986 0.9301

FN FP TN TP ACC SENS SPEC PREC
Freq 2892 5308 926823 33386 0.9915 0.9203 0.9943 0.8628

cytotoxic Tcells-CD8+

plot of chunk unnamed-chunk-59

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9485 0.9527 0.0259 0.9011 0.9493

plot of chunk unnamed-chunk-61

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9343 0.9315 0.0324 0.847 0.9204

B cells (CD3- CD19+)

FN FP TN TP ACC SENS SPEC PREC
Freq 531 895 937430 15167 0.9985 0.9662 0.999 0.9443

FN FP TN TP ACC SENS SPEC PREC
Freq 521 8506 2738594 23528 0.9967 0.9783 0.9969 0.7345
  • typically over- or under-calling the sneaky Bcells on CD19 dimension

B cells (CD3- CD19+)

plot of chunk unnamed-chunk-65

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9377 0.9254 0.0296 0.8679 0.9262

plot of chunk unnamed-chunk-67

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.8809 1.002 0.0428 0.7857 0.8714

naive Bcells (CD27- IgD+)

plot of chunk unnamed-chunk-69

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.909 0.8173 0.0403 0.7344 0.8424

plot of chunk unnamed-chunk-71

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9128 0.8498 0.0367 0.7828 0.868

IgD- memory Bcells (CD27+)

plot of chunk unnamed-chunk-73

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.823 1.16 0.0443 0.8211 0.8764

plot of chunk unnamed-chunk-75

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.8057 0.4608 0.0449 0.4145 0.5765

IgD+ memory Bcells (CD27+)

plot of chunk unnamed-chunk-77

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.9212 0.7128 0.0113 0.964 0.925

plot of chunk unnamed-chunk-79

N_SAMPS N_DATA RHO BETA BETA_SE R2 ICC_1
151 151 0.8844 0.7361 0.0349 0.7487 0.8423

OpenCyto Summary

  • Seems to work
  • Lots of room for improvement
  • Iterative process

Data QC ?

Tested FlowAI

… consists of three key steps to check and remove suspected anomalies that derive from (i) abrupt changes in the flow rate, (ii) instability of signal acquisition and (iii) outliers in the lower limit and margin events in the upper limit of the dynamic range. flowAI: automatic and interactive anomaly discerning tools for flow cytometry data. Monaco, Chen, Poidinger, et al. (2016)

Strange Flowrate Example

Normal Flowrate Example

ICC

From Wikipedia

Cicchetti (1994) gives the following often quoted guidelines for interpretation for kappa or ICC inter-rater agreement measures:

  • Less than 0.40—poor.
  • Between 0.40 and 0.59—Fair.
  • Between 0.60 and 0.74—Good.
  • Between 0.75 and 1.00—Excellent.